250-year-old research methodology helps solve 21st Century population questions

Researchers from the ESRC Centre for Population Change at the University of Southampton and Statistics New Zealand have published an article in the Routledge journal Population Studies arguing that Bayesian methodology, a statistical tool introduced by Rev. Thomas Bayes in the 18th Century, is vital in providing solutions to many difficult statistical problems, particularly those presented by 21st Century population studies.

The open access article examines the achievements of Bayesian methodology and makes the case for its wider use in demography and other population sciences. The results show that there are a number of distinct features of demography that make it especially suited for the application of Bayesian methods and, likewise, demographic studies have a lot to offer back to Bayesian methodology.

Bayesian statistics is unique in its coherent use of probability distributions to describe uncertainty, and the authors, Dr Jakub Bijak and Dr John Bryant, have been pivotal in reviving its recent use in demography. Bayes' theorem was presented to the Royal Society just over 250 years ago, but it was largely unused until recent decades. Its use is now growing fast, following methodological and computational developments, and there is a trend towards probabilistic rather than deterministic perspectives. The United Nations (UN) population projections, for example, have become more probabilistic – and Bayesian – in recent years.

In their article, 'Bayesian demography 250 years after Bayes', Bijak and Bryant argue that Bayes's approach complements many traditional methods which can be re-expressed in Bayesian terms, and that it can coherently bring together information from multiple data sources, incorporating extra information, constraints and expert knowledge, and dealing with incomplete or messy data. Crucially, they argue that Bayesian methodology should be seen as a more general framework, incorporating new insights and responding to new scientific challenges.

The high policy relevance of demographic change gives statisticians working in this area a unique opportunity for applied experimentation with user engagement, communication of uncertainty, and public understanding of statistics. The article demonstrates that a Bayesian methodology framework can incorporate these different aspects, including uncertainty, estimates, and forecasts, and inform policy decisions and analysis of the possible consequences of these decisions.

Dr Bijak comments: "We think that these features of Bayesian statistics, particularly Bayesian demography, are really remarkable for a 250-year-old invention.

"Through our investigations we can see the promise of many further exciting developments in the applied population sciences for years to come. To ensure that Bayesian demography continues to gather momentum, we hope that our research highlights the need for more methodological training opportunities for demographers and social statisticians, an improvement in Bayesian computational methods, and better communication to practitioners around uncertainty and probability distributions for policy and planning purposes."